Rumor: Anthropic's internal Model 4 may have cracked the Riemann Hypothesis; METR chart shows benchmark gains accelerating 1.9x

haider1 · x · 2026-11-21

haider claims Anthropic's internal Model 4 (or an early Model 5 checkpoint) may have solved the Riemann Hypothesis, suggesting internal capabilities far beyond public releases — unverified. The attached METR AECI chart shows benchmark scores growing 1.9x faster (95% CI 1.5x–2.8x) since Oct 2025 when internal models are included: internal Model 4 scores 178.6 vs top public Opus 5.5 at 174.6 (point estimates; Model 4's CI spans 171–193), with annual gain jumping from +15/yr to +29/yr. Author marvels that frontier models can now attack problems that stumped mathematicians for centuries.

Original post →

More from Models

Models channel →